Stop VIO waiting on the LiDAR buffer in LIVO mode

[bing8deng]

Problem

sync_packages() rejects every call up front when
lid_raw_data_buffer is empty, before the mode-specific switch
below even runs:

bool LIVMapper::sync_packages(LidarMeasureGroup &meas)
{
  if (lid_raw_data_buffer.empty() && lidar_en) return false;
  ...

In LIVO mode the LIO and VIO steps alternate, and the VIO step
(the case LIO: branch inside the inner switch) never touches
lid_raw_data_buffer at all — it only reads img_buffer and
imu_buffer. Once LIO has drained the LiDAR queue, this one
precondition still forces VIO to stall until the next LiDAR scan
lands, adding pure gate-wait latency before every VIO update with
no benefit. The stall shrinks as LiDAR rate rises (queue refills
sooner) and grows if LiDAR/camera are hardware-synced at a low
rate, which rules out a "sync harder" fix.

Fix

Move the LiDAR-buffer check out of the shared preamble and into
the two switch branches that actually consume the buffer: the
ONLY_LIO case and the LIVO WAIT/VIO case that performs the
LIO step. The LIVO case that performs the VIO step is left
without the check, since it has no dependency on
lid_raw_data_buffer.

Impact

Measured on our platforms this removes on the order of tens of
milliseconds of gate wait per VIO update at 10 Hz LiDAR input.
No behavior change for ONLY_LIO/ONLY_LO modes or for the LIO
half of LIVO mode.

[bing8deng]

@xuankuzcr @XW-HKU Could you please review this patch?